Volume 81, Issue 6, Pages 1401-1416 (March 2014) Deliberation and Commitment in the Premotor and Primary Motor Cortex during Dynamic Decision Making  David Thura, Paul Cisek  Neuron  Volume 81, Issue 6, Pages 1401-1416 (March 2014) DOI: 10.1016/j.neuron.2014.01.031 Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 1 Task and Behavior (A) The token task (see text). (B) Temporal profile (thick gray line) of the probability that a given target is correct, computed using Equation 1. The vertical, dashed red line indicates the estimated time of the decision (see text), allowing estimation of the success probability at that moment (horizontal, dashed red line). (C) Profiles of success probability of one easy (blue), one ambiguous (green), and one misleading trial (red). A trial is considered easy if the SP exceeds 0.6 after two token jumps and 0.75 after five. A trial is ambiguous if SP is 0.5 after two jumps, between 0.4 and 0.65 after three, and then between 0.55 and 0.66 after five and seven jumps. A trial is misleading if SP is below 0.4 after three jumps. (D) Distributions of decision durations in easy (blue), ambiguous (green), and misleading (red) trials for monkey S (top) and Z (bottom). Shaded regions indicate error trials. Vertical dotted lines indicate the mean (in ms) for each trial type. (E) Cumulative distributions of success probabilities at decision time in the same trial types. Vertical dotted lines indicate mean success probability. (F) Mean (±SE) of the estimated confidence (SumLogLR) at which the decision was made, averaged across all trials grouped as a function of decision duration. Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 2 Neural Activity Tracks the Changing Evidence (A) Response of a decision-related PMd neuron in the tokens task during easy (left), ambiguous (middle), and misleading (right) trials in which the monkey correctly chose the cell’s PT (colored) or OT (gray). Activity is aligned on the first token jump (squares) and truncated 300 ms before movement onset (diamonds) to avoid averaging artifacts. Rasters are sorted by decision duration. (B) Same as A for an example M1 neuron. (C) Top shows the success probability of the PT during easy (blue), ambiguous (green), and misleading (red) trials, in which the monkey correctly chose the PT (solid lines) or OT (dotted lines). Middle shows the average activity of 68 spatially tuned PMd neurons during those same trials. Bottom shows the average activity of 31 spatially tuned M1 cells. Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 3 Evolution of the Relationship between Neural Firing and Sensory Evidence (A) Analysis for one example PMd neuron. Each line illustrates the relationship between the SumLogLR with respect to the PT and the mean neural activity calculated 200 ms later in a 200 ms epoch, color coded from the darkest (first token jump) to the lightest. Only epochs preceding our estimate of DT are included. (B) Same analysis averaged across 68 PMd cells. (C) Same as (A) for an example M1 cell. (D) Same analysis averaged across 31 M1 cells. Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 4 Neural Activity Does Not Integrate the Sensory State (A) Top shows the success probability with respect to PT during bias-up/down (green) and bias-down/up trials (magenta). Bottom shows the activity of a PMd neuron during the two trial types. Only trials in which DT ≥800 ms are included. (B) Top shows the mean success probability during bias-up/down (green) and bias-down/up trials (magenta). Middle shows the average activity of 68 decision-related PMd neurons during the two trial types. Bottom shows the average activity of 31 M1 neurons. (C) Comparison of mean neural activity (±SE) of 68 PMd (top) and 31 M1 neurons (bottom) recorded during bias-up/down versus bias-down/up trials from 600–800 ms after the first token jump (left shaded area in [B]). Colored crosses illustrate neurons with a significant modulation of activity (green indicates stronger activity in bias-up/down; magenta indicates stronger activity in bias-down/up). Percentages denote the proportion of significantly modulated cells. (D) Same as (C), but showing activity from 900–1,100 ms after the first token jump (right shaded area in [B]). Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 5 The Sensory State Is Quickly Tracked (A) Left shows the following six trials for testing integration versus tracking: bias-up/down (green) and bias-down/up (magenta), long bias-for (blue), long bias-against (red), short bias-for (black), and short bias-against (gray). Right shows the cumulative DT distributions in these six trial types, including only trials in which the decision was made after the bias (gray areas in the left panel). (B) Left shows step-for trials (black), in which the first token jumps to the PT and the second to the OT, and step-against trials (red), in which the pattern is reversed. Right shows average (±SE) neural activity in PMd and M1 from 200–400 ms and 400–600 ms after the first token jump. Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 6 Neural Correlates of the Moment of Commitment (A) Response of a tuned PMd neuron in easy (left), ambiguous (middle), and misleading (right) trials. Same convention as in Figure 2A except that here activity is aligned on movement onset (diamonds). Triangles mark movement offset. (B) Same as (A) for an example M1 neuron. (C) Success probability (top) and average activity of 68 PMd (middle) and 31 M1 (bottom) neurons, aligned on movement onset. The vertical gray line indicates our estimate of commitment time (see Figure 7 and text for details). Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 7 Timing and Robustness of the Neural Activity Peak (A) Left shows average activity of the 68 decision-related PMd cells across all trials (in 10ms bins) for PT (solid) and OT choices (dotted). The peak of the average PT-related activity is indicated and its timing defined relative to movement onset. Right shows cumulative distribution of PT- (solid) and OT- related (dotted) peak activity timing relative to movement onset, calculated individually in each of the 68 PMd cells. (B) Relationship between latencies of peak activity of the 68 PMd cells and reaction time. For each cell (black lines), the mean peak activity latencies are calculated in easy (blue) and misleading (red) trials for the cell’s PT and plotted against the mean RTs in these same trials. The inset shows the distribution of slopes of the black lines (dashed line is unity slope). (C) Comparison of the mean firing rate ±SE at commitment time (280 ms before movement onset) of each of the 68 PMd cells during easy-versus-ambiguous (left), easy-versus-misleading (middle), and ambiguous-versus-misleading trials (right). Colored crosses illustrate neurons for which the difference is significant (KS test, p < 0.05). Percentages denote the proportion of neurons whose activity is significantly different in each comparison. (D) Same as (A) for a population of 31 decision-related M1 neurons. (E) Same as (B) for the 31 M1 neurons. (F) Same as (C) for the 31 M1 cells (note that the mean firing rate ±SE of M1 cells is calculated 280 ms before movement onset). Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 8 Comparison of Volitional Commitment versus Instructed Movement (A) Activity of a PMd decision-related neuron during the delayed reach task, in trials to the PT (blue) or OT (gray), aligned on target onset (left), GO signal (middle), and movement onset (right). (B) Activity of the same neuron during easy trials in the tokens task and aligned on movement onset. (C) Average activity of 58 PMd neurons recorded during the DR task, in trials to the PT (blue) or OT (gray). Among these, 30 are significantly tuned in the DR task (solid) whereas 28 are not (dotted). (D) Average activity of the same PMd neurons during easy trials in the tokens task. (E) Same as (A) for an example M1 neuron. (F) Same as (B) for the same M1 neuron as (E). (G) Same as (C) for 18 M1 neurons tuned in the DR task. (H) Same as (D) for the same 18 M1 neurons. Neuron 2014 81, 1401-1416DOI: (10.1016/j.neuron.2014.01.031) Copyright © 2014 Elsevier Inc. Terms and Conditions